R Notebook

Whats in V3: GAM models using Humidity, Temperature, PM2.5 and NPIs

Climate variables (temperature, humidity, PM2.5) lagged so that daily cases are predicted using climate variables from n=climate_lag days prior

NPI lagged so that daily cases are predicted using NPI data from n=0,…,num_iter days prior

the climate splines should be more or less the same since we’re keeping climate_lag constant i think what we’re looking for is an appropriate lag for npi which should change a lot over the num_iter GAMs

If you run this you’ll see that infections decrease with temperature and increase with PM2.5 Humidity is kind of a wildcard. This is overall true for all the cities (florence is a little weird) and for all the values of climate lag I tried (3,4,5,6)

We start by loading in our data tables

require(mgcv)
bologna <- read.csv("bologna_full.csv")
bologna <- na.omit(bologna)

modena <- read.csv("modena_full.csv")
modena <- na.omit(modena)

florence <- read.csv("florence_full.csv")
florence <- na.omit(florence)

rome <- read.csv("rome_full.csv")
rome <- na.omit(rome)

num_iter <- 30
climate_lag <- 6

Bologna:

for(i in 0:num_iter) {
num_lag <- i
bologna <- read.csv("bologna_full.csv") #Read in file
#Push temperature&humidity data back climate_lag days by adding empty data in front
#rep(NA, n) creates a vector of length n filled with NA
bologna$Temperature.1 <- c(rep(NA, climate_lag),  bologna$Temperature[(1):(nrow(bologna)-climate_lag)]) 
bologna$Humidity.1 <- c(rep(NA, climate_lag), bologna$Humidity[(1):(nrow(bologna)-climate_lag)])
bologna$pm25.1 <- c(rep(NA, climate_lag), bologna$pm25[(1):(nrow(bologna)-climate_lag)])


#push npi back num_lag days
bologna$npi.1 <- c(rep(NA, num_lag), bologna$npi[(1):(nrow(bologna)-num_lag)])
#Omit the new NAs
bologna <- na.omit(bologna)

infections.gam <- gam(Daily ~ s(Temperature.1) + s(Humidity.1) + s(pm25.1) + s(npi.1, k=4), family=poisson(link="log"), data=bologna)
plot(infections.gam,scale=0,se=2, shade=TRUE,pages=1)
title(main = i)
}

Modena:

for(i in 0:num_iter) {
num_lag <- i
modena <- read.csv("modena_full.csv") #Read in file
#Push temperature&humidity data back climate_lag days by adding empty data in front
#rep(NA, n) creates a vector of length n filled with NA
modena$Temperature.1 <- c(rep(NA, climate_lag),  modena$Temperature[(1):(nrow(modena)-climate_lag)]) 
modena$Humidity.1 <- c(rep(NA, climate_lag), modena$Humidity[(1):(nrow(modena)-climate_lag)])
modena$pm25.1 <- c(rep(NA, climate_lag), modena$pm25[(1):(nrow(modena)-climate_lag)])

#push npi back num_lag days
modena$npi.1 <- c(rep(NA, num_lag), modena$npi[(1):(nrow(modena)-num_lag)])
#Omit the new NAs
modena <- na.omit(modena)

infections.gam <- gam(Daily ~ s(Temperature.1) + s(Humidity.1) + s(pm25.1) + s(npi.1, k=4), family=poisson(link="log"), data=modena)
plot(infections.gam,scale=0,se=2, shade=TRUE,pages=1)
title(main = i)
}

NA

Florence:

for(i in 0:num_iter) {
num_lag <- i
florence <- read.csv("florence_full.csv") #Read in file
#Push temperature&humidity data back climate_lag days by adding empty data in front
#rep(NA, n) creates a vector of length n filled with NA
florence$Temperature.1 <- c(rep(NA, climate_lag),  florence$Temperature[(1):(nrow(florence)-climate_lag)]) 
florence$Humidity.1 <- c(rep(NA, climate_lag), florence$Humidity[(1):(nrow(florence)-climate_lag)])
florence$pm25.1 <- c(rep(NA, climate_lag), florence$pm25[(1):(nrow(florence)-climate_lag)])

#push npi back num_lag days
florence$npi.1 <- c(rep(NA, num_lag), florence$npi[(1):(nrow(florence)-num_lag)])
#Omit the new NAs
florence <- na.omit(florence)

infections.gam <- gam(Daily ~ s(Temperature.1) + s(Humidity.1) + s(pm25.1) + s(npi.1, k=4), family=poisson(link="log"), data=florence)
plot(infections.gam,scale=0,se=2, shade=TRUE,pages=1)
title(main = i)
}

Rome:

for(i in 0:num_iter) {
num_lag <- i
rome <- read.csv("rome_full.csv") #Read in file
#Push temperature&humidity data back climate_lag days by adding empty data in front
#rep(NA, n) creates a vector of length n filled with NA
rome$Temperature.1 <- c(rep(NA, climate_lag),  rome$Temperature[(1):(nrow(rome)-climate_lag)]) 
rome$Humidity.1 <- c(rep(NA, climate_lag), rome$Humidity[(1):(nrow(rome)-climate_lag)])
rome$pm25.1 <- c(rep(NA, climate_lag), rome$pm25[(1):(nrow(rome)-climate_lag)])

#push npi back num_lag days
rome$npi.1 <- c(rep(NA, num_lag), rome$npi[(1):(nrow(rome)-num_lag)])
#Omit the new NAs
rome <- na.omit(rome)

infections.gam <- gam(Daily ~ s(Temperature.1) + s(Humidity.1) + s(pm25.1)+s(npi.1, k=4), family=poisson(link="log"), data=rome)
plot(infections.gam,scale=0,se=2, shade=TRUE,pages=1)
title(main = i)
}